1 Introduction

This document shows the prediction maps for the possible infection risk of trees in the Basque Country by the following pathogens:

• Armillaria mellea
• Diplodia sapinea
• Fusarium circinatum
• Heterobasidion annosum

The following algorithms were used to create the predictions:

• Boosted Regression Trees (BRT)
• Generalized Additive Model (GAM)
• Generalized Linear Model (GLM)
• k-Nearest Neighbor (KNN)
• Random Forests (RF)
• Support Vector Machine (SVM)
• Extreme Gradient Boosting (XGBOOST)

The following “algorithm-pathogen-resampling” settings did not return a performance result during performance evaluation. Due to the missing performance result for these combinations no prediction maps were created.

Unfortunately, XGBOOST cannot handle new factor levels in prediction data. Since we predict to the whole Basque Country using environment variables, this case occurs quite often. Variables like “soil type” and “lithology type” inherit instances which only occur in some parts of the prediction area but not within the training data. Therefore, it was not possible to create prediction maps for XGBOOST.